Add JWE support

[chrisFrodo]

Add JWE support that is missing currently, and fix a bug with userinfo

[bodewig]

Thanks @chrisFrodo - as you've probably realized this is not the most active project right now. I'm sorry for that.

I've got a few comments and questions, though.

• string_starts is never used
• The way you've added support for userinfo endpoints returning JWTs means openidc_parse_json_response now accepts JWTs for much more than just the userinfo endpoint. This probably should better be conditional.
• openidc_load_jwt_and_verify_crypto already contains code to split the JWT to handle the case where the none alg has been used (only two parts of a JWT). I believe it would be cleaner and easier if we merged JWE support with the code just above - and likely we could then stick to match and destructuring.
• please also update the README with an example as a minimal way of documenting how to configure things when JWEs are used
• why would we want to restrict this to RSA - and only with a very specific key length and only one of the padding alternatives? Wouldn't the same work for other key sizes or even when using AES as well?
• I would love to have tests for this. I know the way testing works is a bit convoluted and it likely takes some effort to make the test user info endpoint return a JWT but most stuff should be there already
• please add yourself to the AUTHORS file

[chrisFrodo]

Hi @bodewig ! I'm sorry for answering this late your questions about this PR, I might have been quite busy on other projects.

Following your suggestion I've made some modification on my previous work :

• Unused function have been removed
• Added a minimal example of the JWE related config options in the README.md file
• Moved the JWS parsing of the userinfo to the openidc.call_userinfo_endpoint()function
• Rework the openidc_load_jwt_and_verify_crypto() function to use the string.match() function. I had to modify the logic of booth the JWS and JWT handling so each case can be handled using the same extracted token parts. I may have added too much comments though...
• Included new unit tests for this PR.

About your question about the cryptographic limitations, the fun begins !
These limitations are mainly linked to the lua-resty-jwt library and its own cryptographic constrains on decryption. The current version of the library (v0.2.3) only support the following settings for JWE tokens :

• Key management algorithm (ie : "alg") :
  • RSA-OAEP-256
  • DIR (RFC non-compliant value) : This value is used to tell the parse_jwe() function is used to directly a pre-shared secret to derivate the decryption key
• Content encryption algorithm (ie : "enc") :
  • A256GCM
  • A128CBC_HS256
  • A256CBC_HS256

So, as described in the OpenID Connect CORE spec, the jwt lib is not complient on the encryption/decryption aspect. It is in fact lacking the support of the RSA1_5 algorithm and should allow non signed version of the CBC mode encryption algorithms.

Here are a list of the algorithm that should be supported, the bold ones being mandatory :

alg	enc
RSA1_5	A128CBC
RSA-OAEP	A256CBC
EDCH-ES	A128GCM
A128KW	A256GCM
A256KW
A128GCM
A256GCM

Fun fact : the HMAC signature of content encrypted with CBC mode is ment be optional

That means some modifications will have to be done on the lua-resty-jwt library in order to meet the minimun requirement of the OpenID Connect spec. I'll try to work on it this month.

Hope these answers helped you!

[bodewig]

Thanks a lot @chrisFrodo , this looks great

apart from the small comment I left inline I wonder whether we need to do something about elliptic curves as well - see also #457 .

The EC MR makes me wonder whether it is a good idea to encode lua-resty-jwt's current limitations in lua-resty-openidc as it will always require a new release here if lua-resty-jwt learns new tricks that may be relevant to users of the library.